Modern wind turbines are electromechanical systems with about 30,000 components built into different subsystems like the rotor, gearbox, generation and transmission. Displacement, wear-tear and failure of these components affects performance of the wind turbine to varying degrees right from small production losses to component failures.
Large number of components adds difficulty to the diagnosis process in event of failures. For the purpose of diagnosis and optimizing maintenance activity, modern wind turbines are equipped with sophisticated Supervisory
Control And Data Acquisition (SCADA) systems. The most advanced of such sensors are displacement sensors to measure position and accelerometers to measure vibration of components. This data is typically continuous in nature and is sampled by the system at an appropriate sampling rate. A stream of discrete data points is generated at this sampling rate and is monitored locally or remotely.
The rate at which data is sampled depends on how fast the data changes which is a function of the subsystem or sub-component under measurement. If a fast changing signal is sampled slowly there would be loss of information and hence the sampling rate is an important parameter.
Most industrial data loggers may sample data at a different rate from the rate at which the data is logged. So while data is sampled every few seconds they are programmed to store output data as 10 minute average values. The impact of the averaging is similar to impact of the lower sampling i.e. information loss. Average values (like shown in the image as the red line) will not trigger an alarm even though spikes in the output that crossed preset tolerances may have occurred several times.
The data stream when visualized as a physical quantity (displacement or vibration) varying over time can be analyzed to get insights into the condition of the different subsystems. For example a higher displacement output can convey the shifting of the gear assembly beyond its prescribed limit or a climbing temperature can convey greater friction than expected. Making deductions from signals based on variations over time is known as Time Domain Analysis.
Another important form of analysis is Frequency Domain Analysis wherein the time domain signal is converted into a frequency domain signal (time and frequency are inversely related F= 1/T) and its component frequencies are analyzed. It was Jean Baptiste Fourier who in the 18th century invented the mathematics for visualizing the frequency spectrum of a time based signal.
So what is the utility of the frequency domain analysis? Fault detection based on variations in frequency is a useful tool in multiple cases. Some examples of faults that can be detected in the frequency domain are
· Shaft deflection
· Bearing displacement
· Weakness or cracks in the foundation
· Loosening of structural components
These faults affect the output of different sensors in specific ways. These patterns may not be visible in the time domain signal but will show up as discrete frequency spikes in the frequency spectrum. We have been able to ascertain and label frequency spikes to their associated turbine issues.
Algo Engines enables time and frequency domain analysis of sensor data from wind turbines. The ability to capture and log data at a high frequency as well as efficient fast Fourier transform (FFT) algorithms enables end users to identify deviations from standard operating levels. For more on frequency domain analysis, read our future posts on this blog.